In the past, a wavefront synthesis technology is known which collects sound wavefront using a microphone array formed with a plurality of microphones in sound collection space and reproduces sound using a speaker array formed with a plurality of speakers on the basis of obtained multichannel sound signals. Upon reproduction of sound, sound is separated as necessary so that only sound from a desired sound source is reproduced.
For example, as a sound source separation technology, a minimum variance beam former, multichannel nonnegative martrix factorization (NMF) (nonnegative matrix factorization), or the like, which estimate a time-frequency mask using an inverse matrix of a microphone correlation matrix formed with elements indicating correlation between microphones, that is, between channels, are known (for example, see Non-Patent Literature 1 and Non-Patent Literature 2).
By utilizing such a sound source separation technology, it is possible to extract and reproduce only sound from a desired sound source using a time-frequency mask.